LLM from Scratch, No Slide, Just Code

Large Language Models are everywhere, but how they actually work “under the hood” often remains a mystery. In this live-coding talk, we’ll skip the slides and build an LLM step by step, not just a toy model, but a state-of-the-art one. Along the way we’ll dive into tokenization, transformer architecture, small-scale training, and inference, showing how these components come together to power today’s most advanced models.